hybrid location-based recommender system for...
TRANSCRIPT
Hybrid Location-based Recommender System for Mobilityand Travel Planning
Logesh Ravi1 & V. Subramaniyaswamy1 & V. Vijayakumar2 & Siguang Chen3& A. Karmel2 & Malathi Devarajan1
# Springer Science+Business Media, LLC, part of Springer Nature 2019
AbstractIn recent times, the modern developments of internet technologies and social networks have attracted global researchers toexplore the recommender systems for generating personalized location-based services. Recommender Systems (RSs) as provendecision support tools have gained immense popularity to solve information overloading problem among various real-timeapplications of e-commerce, travel and tourism,movies and e-learning. RSs emerge as a popular and reliable information filteringapproach that is capable of suggesting relevant items, movies, and locations to the active target user based on dynamic prefer-ences and interests. Beyond the development of many feature-rich recommendation algorithms, the need for a better full-fledgedRS to produce precise and highly relevant recommendations based on ratings and preferences provided by the target user is veryhigh. With the specific focus to the travel domain, the global research community has been involved in the development of acomplete travel recommender system that is immune to the sparsity and cold start problems. In this paper, we present a newHybrid Location-based Travel Recommender System (HLTRS) through exploiting ensemble based co-training method withswarm intelligence algorithms to enhance the personalized travel recommendations. The proposed HLTRS is experimentallyvalidated on the real-world large-scale dataset, and we have made an extensive user study to determine the ability of developedRS to produce user satisfiable recommendations in real-time scenarios. The obtained results and analyses demonstrate theimproved performance of the proposed Hybrid Location-based Travel Recommender System over existing baselines of recom-mender systems research.
Keywords Recommender systems . Point-of-interest . Travel recommendation . Personalization . Travel planning
1 Introduction
Recommender System (RS) has emerged as an effective deci-sion support mechanism to tackle information overloadingproblem faced by digital users during the search of relevantinteresting items that meet their changing requirements in re-cent years [1]. In various real-time applications such asmovies, music, books, tourism, e-commerce, and education,
Recommender Systems are widely used [2–7]. Among vari-ous techniques developed for making personalized recom-mendations, Collaborative Filtering, Content-based Filtering,and Hybrid Filtering were widely utilized to support activetarget users by providing a personalized list of recommenda-tions. The rapid increase in the utilization of Smartphones andsocial networks has resulted with the huge amount of user-generated data, and this data is considered to be a significant
* V. [email protected]
Logesh [email protected]
Siguang [email protected]
Malathi [email protected]
1 School of Computing, SASTRA Deemed University,Thanjavur, India
2 School of Computing Science and Engineering, Vellore Institute ofTechnology, Chennai, India
3 Key Lab of Broadband Wireless Communication and SensorNetwork Technology of Ministry of Education, Nanjing Universityof Posts and Telecommunications, Nanjing, China
Mobile Networks and Applicationshttps://doi.org/10.1007/s11036-019-01260-4
source for the recommender systems to produce personalizedrecommendation list [8]. Figure 1 depicts the overview of therecommendation generation process in the recommender sys-tems. Online users share their thoughts through social networkand recommendations are presented by considering targetusers’ social behaviors and activities.
The growing development of web-based technologies andsocial networks, users tend to share their opinions andthoughts online [9]. As location-based social network helpsto share their experiences, it is essential to built users’ profilebased on social activities and behavior to learn more aboutusers’ preferences from historical check-ins [10, 11]. Travelrecommender system suggests point of interests to the user bymining users’ interests and preferences implicitly or explicitlyfrom location-based social networks. Based on users’ his-torical check-ins, similar users are identified and rating ispredicted for the unknown places. To provide personalizedrecommendations in real-time applications, a good recom-mender system should consider user requirements, socialties, geographical features and contextual information topredict highly relevant items which are more probably tobe accepted by the target user [12]. Though collaborativefiltering based recommender system alleviates informationoverloading problem, they are still suffering from coldstart, sparsity and personalization problems. To addressthe aforementioned issues, most researchers have usedopinions from trusted users to generate personalized rec-ommendations rather than from similar users [13].
In collaborative filtering based recommender system, dif-ferent clustering algorithms were used to discover similarusers. But, the traditional clustering algorithm has some draw-backs to yield optimal solutions [14]. To conquer such limita-tions, swarm intelligent algorithms have been introduced foreffective user clustering. The emergence of swarm intelli-gence algorithms has attracted many researchers to exploitsuch algorithms in recommender system for user clustering[15]. The nature-inspired meta-heuristic algorithms such as
cuckoo search, particle swarm optimization and ant colonyoptimization algorithms were utilized in various applicationsto tackle complex optimization problems [16–18]. Bio-inspired algorithm inherits the features of the biological sys-tem and yields promising convergence results [19]. Especiallyin travel recommendations, both the number of travellers andpoint of interests are large, but user rating is very low whichleads to data sparsity problem. Hence correlation betweentravellers cannot be derived appropriately, and so it may notbe reliable. Similarly, the problem of cold start arises whentravel recommender system tries to generate recommenda-tions for the unfamiliar point of interests or for new travellers[20]. Thus the recommender system gains a negative impres-sion on the performance of traditional filtering algorithms. Itwill be more critical to generate personalized recommenda-tions as it contains only very few rating information.
Recently, Da Costa et al. [21] proposed Co-trainingbased recommender system (CoRec) to resolve cold startand sparsity problem by strengthening the rating matrixas a pre-processing step. In 1998, Blum and Mitchellintroduced the concept of Co-training as a distinctmulti-view learning model [22]. Co-training algorithmuses multiple views of a single feature set that are uncorrelatedby its description and compatible if all are identically labeled.As an extension ECo-Rec is proposed, where CoRec is gen-eralized to work with more than two recommendation algo-rithms and aggregate the results of simultaneously co-trainedmodels to generate consensus user-item rating matrix [23].Emerging demand for personalized travel recommendationsbroadens the scope for the significant progress of location-based travel recommender system. We have utilized the co-training method in the travel recommendation process to han-dle user-POI sparsity and cold start problem of traditionalrecommender system by predicting ratings for unfamiliarPOIs and new user. The proposed HLTRS is validated onthe TripAdvisor dataset to demonstrate the applicability ofthe proposed travel recommendation generation process in
Fig. 1 Overview of therecommender system andrecommendation generationprocess
Mobile Netw Appl
real time scenarios. The major contributions of the proposedHLTRS are summarized as follows.
& We have developed a personalized travel recommendersystem by analyzing target users’ online social activitiesand behaviors.
& We have employed ensemble based co-trained swarm in-telligence algorithms to identify the best neighborhood forrating prediction and to enhance the rating matrix.
& Contextual information has been utilized to minimize thecomputational complexity and dimensionality reductionof the recommendation generation process.
& The proposed HLTRS is experimentally validated on thereal time complex dataset to exhibit the performance andefficiency over existing baselines.
The remainder of the article is structured as follows: Insection 2, recent literature related to travel recommender sys-tem, collaborative filtering algorithms, social user profiling,swarm intelligence for effective user clustering and co-training method for rating enhancement are presented. In sec-tion 3, the detailed explanation of the proposed HybridLocation-based Travel Recommender System (HLTRS) ispresented. Section 4 portrays the results obtained from exper-iments and discusses the performance efficiency of the pro-posed HLTRS over other existing recommendation models.Finally, in section 5, we conclude the work done along withfuture research directions.
2 Related works
Recently, the recommender system has been widely exploredin various application domains with different methodologiesto strengthen the performance efficiency of the recommenda-tion generation process. The overview of the conventionalrecommender system framed using collaborative filteringtechniques; advantages of using swarm intelligent algorithmsin the recommender system, trust-aware recommendations,and co-training method for rating enhancement are presentedin the following subsections.
2.1 Recommender systems based on collaborativefiltering
The most generally used filtering algorithm in the recom-mender system is collaborative filtering. Collaborative filter-ing algorithm predicts the rating for unknown items of thetarget user based on the ratings of the items provided by theneighbors [24]. Goldberg et al. [25] first established the ideaof collaborative filtering to discover similar users and now it iswidely used technique in the recommender systems. Later, in1997, GroupLens [26] utilized the collaborative filtering
algorithm to group the news articles automatically in Usenet.Herlocker et al. [27] present the measuring factors to estimatethe quality of the rating prediction mechanism of recommend-er systems. Despite the benefits provided by the collaborativefiltering algorithm, cold start and data sparsity are the majorconcern of the conventional recommender systems. Thoughsome existing collaborative filtering based recommender sys-tems provide better results, at most times the contextual infor-mation and user preferences are not considered to producerecommendations [28]. These collaborative filtering algo-rithms sometimes necessitate domain experts for valuablesuggestions.
2.2 Trust-aware recommendations
As the collaborative filtering based recommender system isvulnerable to malicious attacks [29, 30] and data sparsityproblem, incorporating trust information helps to conquersuch issues. This is due to the fact that people used to buyitems suggested by friends, family members and colleagueswhom they believe the most. Lathia et al. [31] have proposedthe collaborative filtering based recommender systemwith thetrusted k-nearest neighbor to alleviate the problems of thetraditional recommender system. Massa and Avesani [32]used four different trust metrics to alleviate ratings sparsityand cold start problems of the conventional recommender sys-tems. Kant and Bharadwaj [33] exploit both trust and untrustinformation in the recommender system to enhance the rec-ommendation quality through the fuzzy model. Guo et al. [34]performed experimental research on trust metrics of the rec-ommender system and stated that each trust metric has itsadvantages and limitations depend on the context of usage.Gupta and Nagpal [35] present the survey on trust metrics andits properties for their application in various contexts. As rec-ommendation generation process grows, trust metrics need tobe recomputed to produce good results.
2.3 Swarm intelligent algorithms in recommendersystems
Various clustering algorithms were used in collaborative fil-tering based recommender system to discover neighbor setsfor rating predictions. Due to the drawback of the convention-al clustering algorithm to obtain optimal solutions on largescale applications, many researchers have incorporated swarmintelligence algorithms in various recommender systems foruser clustering. Swarm intelligent algorithms inherit the fea-tures of biological systems to provide better convergence rate.Ujjin and Bentley [36] exploit particle swarm intelligent algo-rithm to produce tailored recommendations using a modifiedEuclidean distance. The experiments were conducted onMovieLens datasets, and showed that PSO performs betterthan the genetic algorithm. Katarya and Verma [37] developed
Mobile Netw Appl
a recommendation model by combining particle swarm opti-mization and fuzzy c-means algorithm to generate personal-ized movie recommendations with reduced computationalcost. To find weights for features of Jester dataset,Choudhary et al. [38] utilized Gravitational search algorithm(GSA) and showed that GSA outperforms PSO. Bedi andSharma [39] employed Ant colony optimization (ACO) fortrust computation, and it solves the problem like local optima.Likewise, meta-heuristic algorithms such as Mussels wander-ing optimization (MWO) [40] and Cuckoo search algorithm(CSA) [41] are used to address the limitations of traditionalclustering such as k-means. The integration of k-means withMWO and CSO provides better clustering results than indi-vidual algorithms. In general, every clustering algorithm hasits limitations and thus integrating results of two or more hy-brid clustering algorithms helps to generate promising solu-tions [42, 43].
2.4 Travel recommender system
Travel Recommender system has gained significant interestfrom many researchers due to the employment of swarm in-telligence for user clustering algorithms and personalized rec-ommendation generations [44–46]. Themajor goal of the trav-el recommender system is to satisfy users’ requirements.Various parameters such as user preference, interest, weather,time of the day, companion and travel costs are considered assome constraints while making recommendations. Based onthe target users’ requirements, the influential features of thetravel recommender system are adjusted to generate the rele-vant point of interests. Logesh et al. [47] proposed a recom-mendation model called DPSOHiK for personalized point ofinterest recommendations based on hybrid particle swarm op-timization algorithm using electroencephalography feedback.Brilhante et al. [48] developed a web-based travel recom-mendation model called TripBuilder. They acquired theinformation from Wikipedia and Flickr dataset to evaluatethe performance of TripBuilder. Similarly, Kurata and Hara[49] proposed a travel plan recommender called CT-Planner by exploiting genetic algorithms. Notably, mostof the available travel recommender system facilitates gen-eral recommendations and lack personalization. Logeshand Subramaniyaswamy [50] developed a hybrid travelrecommendation model called PCAHTRS using contextualdata to facilitate personalized recommendations.
2.5 Co-training method
The challenging issue of any recommender system is to dealwith large user-item matrix where most of the entries are emp-ty [51]. One of the most useful ways to handle rating sparsityand cold start difficulty of the conventional recommender sys-tem is to fill those missing values. Recently, some researchers
have utilized the co-training method in the recommender sys-tem to enrich the rating matrix. Zhang et al. [52] presented acontext-aware semi-supervised co-training model calledCSEL to tackle the cold start problem of the recommendersystem. Zhang and Wang [53] developed a unifiedCollaborative Multi-view Learning (CML) model to alleviatethe sparsity problem of recommender system by providing anadditional view to strengthen the sparse user-item rating ma-trix. Quang et al. [54] utilized a co-training based collaborat-ing filtering method that iteratively extends the training set byexchanging user and item features to provide enhanced pre-diction accuracy. More recently, Matuszyk and Spiliopoulou[55] introduced a semi-supervised stream based recommenda-tion model with the co-training approach. Here the predictionmodel generates reliable predictions and uses the same as alabel to improve the recommendation generation process. Toyield enhanced recommendations, the results obtained fromindividual algorithms are joined together as an ensemble togenerate consensus rating matrix for rating prediction [56].
3 Proposed hybrid location-based travelrecommender system
The most important aim of the proposed HLTRS is to makethe search process of the tourist simpler and more convenient.The proposed HLTRS generates POIs and travel routes whichare more specific to the target user. Based on the historicalcheck-ins, interests, preferences, and requirements, personal-ized recommendations are generated. Though lots of travelrecommender systems are available, they suffer from coldstart and data sparsity problem. Despite, existing travel rec-ommender system facilitates only general recommendationsand not user-specific or personalized recommendations.Travel Recommender systems are decision-making tools anddifferent from other recommender systems, analyzing users’interests and preferences are considered as an essential factorto construct efficient user-centric recommender system. Themajor goal of the proposed HLTRS is to generate personalizedrecommendations for travellers by exploiting social networkdata. Information such as historical check-ins, social ties, feed-backs and reviews about already visited sites, social activitiesand behavior are acquired from users’ social network. Bymining such information and building user profiles, it be-comes easier to develop an efficient personalized travel rec-ommendation model.
The proposed HLTRS is designed for travellers to facilitatepersonalized recommendations based on users’ social activi-ties and preferences. Our proposed Hybrid Location-basedTravel Recommender System builds user profile from socialnetwork data, and contextual data are utilized to remove irrel-evant information. To strengthen the user-POI rating matrix,an ensemble based co-trained swarm intelligence method is
Mobile Netw Appl
employed for user clustering, and finally personalized recom-mendations are presented to the target user. The recommenda-tion generation process of the proposedHLTRS is portrayed inFig. 2. Initially, the user profile is constructed to learn abouttarget users’ social activities and behavior to boost the recom-mendation generation process. Based on the profile built fromsocial data, the user is clustered into a group from wherenearest neighbors are selected for rating prediction. InHLTRS, Co-training technique is used to strengthen theuser-POI rating matrix for accurate rating prediction. It signif-icantly reduces the rating sparsity and cold start difficulty ofthe conventional travel recommender system. Then the ratingpredicted POIs are ranked and top ranked POIs are suggestedas the personalized recommendations.
3.1 User profiling
The proposed HLTRS aims to provide personalized travelrecommendations on the basis of users’ interests, prefer-ences, and requirements. To learn more about the prefer-ences of the target user, the user profile is constructed fromthe social activities and behaviors collected from locationbased social network data. User profiles are modeled withrespect to demographic features and preferences.Demographic features consist of users’ personal detailssuch as age, gender, and employment, whereas preferencesconsist of users’ spatial-temporal features such as historicalcheck-ins, reviews, and feedbacks, comments, posts, andrequirements. A users’ preference on POI varies in associ-ation with weather, season of the year, time preferred totravel and with whom they wish to travel like family orfriends, interests and travel cost. For instance, people usu-ally go for sightseeing in the morning and evening, and
have food at noon. Thus, we utilized location-based socialnetwork data to extract the influential features of the usercheck-ins.
The multi-layer user profiling model based on online re-views and ratings are employed in HLTRS. Each layer sym-bolizes the user preference with respect to influential featuresets such as weather, cost estimation, companion, travel fre-quency and time. Only positive reviews of the user feedbackare used to build a user profile. The reason behind choosingonly positive reviews is to discover and suggest the point ofinterests that are more relevant to the users’ interests. If therating provided by the target user is lesser than 3 stars, it isconsidered as a negative review and ignored, while others areconsidered as a positive review. Since traveler used to ratetheir experience in the form of Bfive stars^, they are used topredict the rating for unfamiliar point of interests. The multi-layer user profile θu built by language model [57] is defined asfollows
θu ¼∑ri∈R f
p wjrið ÞjRf j ð1Þ
where Rf = ∪ ri is the union of all positive reviews about thefeature i, w represents the influential word in the review andp(w| ri) represents the conditional probability estimated as fol-lows
p wjrið Þ ¼Friw þ μ
FR fw
∑wFR fw
∑wFriw þ μ
ð2Þ
where μ is the smoothing parameter, Friw is the influential
word w frequency of ri and FR fw is the influential word w
frequency of Rf.
Fig. 2 Recommendationgeneration process of theproposed HLTRS
Mobile Netw Appl
3.2 Trust utilization for better personalization
The users’ rating profile has various features to predict thepoint of interests for the target user. But users’ trust net-work does not provide such features to discover similarneighbor sets of the target user. Thus, measuring the per-formance efficiency of both the rating profile and trustnetworks helps to predict the point of interests more accu-rately. In Hybrid Location-based Travel RecommenderSystem, user reliability is computed to identify the userwith low rating profile and trust network with an intentionto strengthen the rating profile. Larger the number of avail-able ratings, the probability of predicting the relevant pointof interests for that user is also high. This is due to the factthat a large number of similar users is identified and byestimating the similarity score between users, there existsa high probability to generate more relevant POIs. Finally,ratings are predicted more accurately for unfamiliar pointof interests and recommendation generated are more prob-ably accepted by the target user. Similarly, the cardinalityof a neighbor set and the cardinality of ratings provided bythat neighbors also influence the estimation of user reli-ability. The summation of the similarities computed be-tween the target user and similar users helps to strengthenthe performance of the rating prediction algorithms.
3.3 Swarm intelligence for user clustering
With the growing popularity of online users and the web-based services, the need for personalized travel recom-mender system increases. We incorporate swarm intelli-gence algorithms for user clustering and ratings are pre-dicted for the POIs from the consensus similarity matrixobtained from the co-trained swarm intelligent algorithmsfor better travel recommendations. The proposed HLTRSexploits three swarm intelligent algorithms such asMussels wandering, ant colony and Cuckoo search optimi-zation algorithms and for user clustering. The HLTRS isdesigned to cluster the input dataset independently byswarm intelligent algorithms, and then the resulted similar-ity matrix is used as additional information for other swarmintelligent algorithms for user clustering. Thus a swarmintelligent algorithm not only learns from its model butalso from the other algorithms for efficient clustering.The process continues for some iteration to generate betteruser-POI rating matrix. Larger the number of iterationsmeans higher the prediction accuracy. Finally, the similar-ity matrices obtained from co-trained swarm intelligent al-gorithms are combined to generate a consensus similaritymatrix through an ensemble approach [58]. From the con-sensus matrix, ratings are predicted, and top-n POIs arerecommended which are more probably to be accepted bythe target user.
3.3.1 Ant colony optimization
The new swarm intelligent algorithm called Ant colonyoptimization (ACO) [59] was introduced to address theNP-hard optimization problem. It inherits the foragingcharacteristics of an ant colony to obtain the optimal solu-tion in high dimensional search space. Ants try to find theirshortest way to their habitat from the food source throughpheromone as a communicating factor. While moving,each ant deposits some amount of pheromone on its solu-tion component and finds the next solution componentwith some probabilistic rules. The key feature of the antcolony optimization is the combination of posterior infor-mation about the last obtained local optima and prior in-formation about the global optima. ACO is represented bythe weighted graph, and each ant generates some amountof pheromone on the edges to communicate with each oth-er while finding their optimal path. The amount of phero-mone released indicates the quality of the solution obtain-ed. To avoid local optima, the pheromone evaporates aftersome time which helps to explore a new path in the solu-tion space. It significantly reduces the probability ofchoosing a longer path, as ants used to prefer the shortestpath that contains a higher amount of pheromone.Considering the better exploration and exploitation featureof the ACO, it is utilized for user clustering.
3.3.2 Cuckoo search algorithm
Cuckoo Search Algorithm (CSA) [60] follows the biologicalfeatures of cuckoo to solve complex optimization problems.CSA inherits the parasitic characteristics of cuckoo which laysits egg randomly on other birds’ nest at a maximum distance.The dissimilar cuckoos’ eggs are identified by the host birdand excluded from the nest; else the chick grows in the hostnest. Otherwise, the host bird disposes its own nest and buildsthe new one somewhere else. Cuckoo used to move from oneplace to another in search of another nest. The breeding andtransition behavior of the cuckoo is employed in data cluster-ing to address the optimization problem of complex datasets.The key advantage of CSA is its simplicity and easy ignoranceof local optima. But its convergence rate is slow and resultingin low clustering accuracy. To enrich the convergence rate andclustering accuracy, the nest finding distance is to be set flex-ible for global optimization. To determine the best global nest,the Markov chain method is utilized. In the Markov chainmodel, the presence of the next nearby nest is determined bythe location of the current nest. The standard CSA may resultin premature convergence, as all cuckoos follow the samesearching method and got trap in the local optima. Thus, thecombination of k-means with cuckoo search algorithm helpsto enrich the convergence rate.
Mobile Netw Appl
3.3.3 Mussels wandering optimization
The new optimization algorithm is developed asMussels Wandering Optimization (MWO) inspired fromthe mussels’ bed formation behavior at complex marinesurfaces to form their habitat. The biological behaviorsof mussels are inherited in MWO to obtain an optimalsolution in clustering problems [40]. Mussels join to-gether for physical protection and move less to maintainthe bed density. However, if the cluster grows greater,mussels decide to move away and forms bed based onRandom Levy walk. Let us assume that the mussel pop-ulation consists of N individuals and forms the bed inthe marine region of d-dimensional space. Let f(s) bethe objective function that represents nutrition. TheMWO algorithm consists of the following six steps toprovide optimal solutions. They are 1) initializing mus-sels population and other parameters, 2) computing bothshort and long range mussel density, 3) establishing thetransition strategy, 4) updating the new position, 5)evaluating the fitness of updated position and 6) exam-ining the terminate criteria. Notably, the bed formationwith Random Levy walk to discover the most appropri-ate place for habitat represents the global optimization.Thus, the optimization algorithm developed from theecological behavior of the mussels helps to provide anoptimal solution.
3.4 Rating enhancement by co-training
In most of the real-time applications, data are describedeither by multiple distinct features or by multiple views ofa single feature. Co-training is related to the problem ofmulti-view learning of the same dataset. In general, Co-training technique is implemented on datasets which con-sist of very few labeled data, and all others are unlabeleddata. The co-training technique first learns the classifierfor each view of labeled data and predictions are madeon unlabelled data to obtain additional information. Theobtained additional information is feed to other classifiersas input and process repeat for some iteration to enrich theclassification accuracy. To provide additional information,each data set should be represented by at least two views.However, co-training can also be performed for a singlerepresentation of the data. To handle rating sparsity andcold start issues of the conventional recommender system,co-training method is exploited in the proposed recom-mendation generation process.
HLTRS uses single dataset as an input and three differentswarm intelligent algorithms as multiple views. The similaritymatrix produced by each clustering algorithm is combined togenerate a consensus matrix for rating prediction. While co-training the clustering algorithms, each algorithm is trained
independently to agree with the clusters of other algorithms,so that they learn not only from its own approach but also fromother algorithms. At each step of the co-training module, theswarm intelligent based user clustering algorithm generatesthe similarity matrix, and it is added to the training datasetof other clustering algorithm and repeats the process iterative-ly. This additional information is exchanged over all other userclustering algorithms iteratively to enrich the rating matrix. Inaddition, the results of each clustering algorithm are combinedto generate a consensus rating matrix for rating prediction andprecise recommendation generation.
In this module, the HLTRS divides the original inputdataset into the rated dataset rlu and unrated dataset ulu .In the rated dataset, the user-POI matrix is filled withratings i.e. all the entries of user-POI rating matrix isfilled, while unrated dataset consists of randomly select-ed users and POI. Since user-POI rating matrix is sparseand unrated dataset is large, including all the unrateduser and POI into the prediction algorithm resulting inslow convergence. To reduce the burden of the co-training process, x number of user and POIs per user israndomly selected. The goal of the co-training module isto predict the rating for unvisited POI. The user similar-ities are computed to predict the ratings for unfamiliarPOIs based on similar users’ rating profiles. In HLTRS,Pearson coefficient is applied to estimate the similarityscore and it is defined as follows:
sim u1; u2ð Þ ¼∑l∈Su1 ;u2
rlu1−ru1� �
rlu2−ru2� �
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi∑l∈Su1 ;u2
rlu1−ru1� �2
r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi∑l∈Su1 ;u2
rlu2−ru2� �2
r ð3Þ
where Su1;u2 is the set of POI visited and reviewed by both the
users u1 and u2, rlu1 is the POI l rated by the user u1 and rlu1 is the
average of all POI ratings provided by u1. Similarly, rlu2 is the
POI l rated by u2 and rlu2 is the average of all POI ratings of u2.Then the rating prediction for unfamiliar POI of the user u1 isgiven as follows
Pu1;l ¼ ru1 þ∑ui∈ku1 ;l
sim u1; uið Þ rui;l−rui� �
∑ui∈ku1 ;lsim u1; uið Þ ð4Þ
where ru1 is the average of all POI ratings given by u1, ku1;l isthe k-nearest neighbors of the user u1 who already rated the POIfor which the rating is going to predicted on behalf of the useru1, rui;l is the rate of the POI l given by the user ui and sim(u1,ui) is the similarity measurement between the user u1 and ui.After predicting the ratings for unvisited POI of the use u1, thetop-n rated POIs are suggested to the user as personalized rec-ommendations. If the target user is not convinced with therecommended POIs, then the list is revised again and generatedbased on his requirements and preferences.
Mobile Netw Appl
3.5 Incorporating contextual information for bettertravel recommendations
Travel Recommender system utilizes the experiences andopinions of the similar user of a community to recommendthe point of interest from the set of available choices. Theproposed HLTRS integrates the contextual information withuser profiles to maximize performance efficiency. The contex-tual data such as season, social ties, weather, time and dynam-ic feature of target user i.e. emotions are considered as animportant feature for travel recommender system. Accordingto Adomavicius and Tuzhilin [61], contextual information canbe incorporated either before or after applying the recommen-dation algorithm to filter the relevant items from the master setof choices. In the contextual pre-filtering model, contextualdata are first utilized to filter irrelevant items and then therecommendation algorithm is applied for recommendationgeneration. But in the post-filtering method, the recommenda-tion algorithm is first applied to the raw dataset, and then thecontextual information is utilized to update the recommendedlist based on user preferences and requirements.
The proposed HLTRS incorporates a contextual pre-filtering method to filter out irrelevant information resultingin reduced computational complexity. The distance betweenthe users’ current location and POI is estimated using geo-graphical features such as latitude and longitude. Then the listof point of interests located within the predefined radius issuggested to the target user based on their interests.Collaborative filtering algorithm builds the user-POI matrixfrom user profiles where each entry represents the ratings pro-vided by the target user for the corresponding POI. Let usconsider the rating matrix with m users, where U = {u1, u2,⋯, um} and n POI, where L = {l1, l2,⋯, ln}. Each user wouldhave experienced some set of POIs, and they would haveexpressed their feedback as a 5-star rating. By computingthe similarity between users, collaborative filtering algorithmidentifies the neighbor set of same interest and predicts therating for the unfamiliar POIs of the user. Then the list oftop-n ranked location is recommended as the point of interest.
4 Experiments and discussions
The proposed Hybrid Location-based Travel RecommenderSystem (HLTRS) is experimentally evaluated to demonstrateits efficiency and performance in generating a personalized listof Point of Interests. We conducted the experiment on realtime complex TripAdvisor dataset for two different finalNumber of Recommendations (NR) as 10 and 20.TripAdvisor is a popular travel recommendation web portalthat comprises of users feedbacks and reviews on the venuesand locations. The dataset used for the experimentation com-prises of 9149 venues, 13,410 users, and 152,721 user ratings
for locations. The dataset is divided into 80% for training and20% for testing purpose. The results obtained from experi-mental study are compared with the other existing baselinemethods such as User-based KNN [26], Item-based KNN[62], Biased MF [63], KBTRS [64], DPSOHiK [47],PCAHTRS [50], and HSSRS [10], and the analyses are pre-sented for the better understanding of the effectiveness ofHLTRS. Experiments are performed on the PC running onthe 64-bit Windows 10 Operating System with Intel Core i7-5500U clocked at 3.00 GHz and 16 GB of memory. Theexperiments for generating recommendations on the mobileframework are conducted on the smartphone with Snapdragon660 Octa-Core clocked at 1.95 GHz with 6 GB of memory.
4.1 Evaluation metrics
The major aim of conducting the experiments is to determinethe ability of the proposed HLTRS for generating recommen-dation of the personalized POIs list to the target user. Weexploit six standard evaluation metrics such as MeanAbsolute Error (MSE), Root Mean Squared Error (RMSE),Coverage, Recall, Precision, and F-Measure to validate theperformance efficiency of the recommendation approaches.The evaluation metrics are defined as follows.
4.1.1 Root mean squared error
Root Mean Squared Error is a popular recommendation eval-uation metric used to estimate the accuracy of the predictedratings, and it is calculated as follows.
RMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi∑user;poi ActualRatingsuser;poi−PredictedRatingsuser;poi
� �2N
sð5Þ
4.1.2 Mean absolute error
Mean Absolute Error is the alternative metric to RMSE usedto compute the difference between the actual and predictedratings, and it is computed as follows.
MAE ¼ 1
Nusers∑
user¼1
Nusers
jActualratings poiuserð Þ−Predictedratings poiuserð Þj ð6Þ
4.1.3 Coverage
Coverage metric is used to determine the percentage of pre-dicted ratings among the total number of ratings consideredfor the experiments. Coverage metric is defined as follows.
Coverage ¼ Number of Predicted RatingsTotal Number of Ratings in Test Set
ð7Þ
Mobile Netw Appl
4.1.4 Precision
Precision is a popular evaluation metric used to determine theprediction and ranking ability of the recommendation ap-proach by exploiting the final recommendations presentedto the active target user. Precision metric considers thenumber of relevant POIs in the presented recommendationlist to compute the positive prediction score and it is cal-culated as follows.
Precision ¼ jRecommendedPOI userð Þ∩RelevantPOI userð ÞjjRecommendedPOI userð Þj ð8Þ
4.1.5 Recall
Recall is also commonly known as the sensitive metric whichis used to compute the usefulness of generated recommenda-tions and it is calculated as follows.
Recall ¼ jRecommendedPOI userð Þ∩RelevantPOI userð ÞjjRelevantPOI userð Þj ð9Þ
4.1.6 F-measure
F-Measure is the harmonic mean of recall and precision met-rics which is used to determine the quality of generated POIrecommendations and it is defined as follows:
F−Measure ¼ 2� Precision� RecallPrecisionþ Recall
ð10Þ
4.2 Comparison of experimental results
This subsection presents the experimental results and analyseswith standard evaluation metrics for the production of person-alized travel recommendations on real time complexTripAdvisor dataset. From the info-graphics Figs. 3, 4, 5, 6,7 and 8, it is apparent that our proposed HLTRS is proficientof producing better recommendation in terms of predictionand raking of recommendations. Recommendations are gen-erated for the final number of recommendations (NR) as 10and 20. With focused designing of HLTRS to produce effi-cient and accurate recommendations, the proposed HLTRShas less RMSE and MAE compared to the existing baselines.On the coverage estimation, the HLTRS is capable ofpredicting maximum user-item pairs in the dataset used forthe experimentation. While considering the precision and re-call, the HLTRS has performed well for the different numberof recommendations. On evaluating the quality of the recom-mendations, the HLTRS has highest F-Measure score overexisting standalone and hybrid approaches. The experimental
result reveals the enhanced performance of the hybrid ap-proaches over standalone models. In terms of coverage,User-based KNN and Item-based KNN have the least capacityto make predictions to produce recommendations, whichmakes these approaches to produce poor quality recommen-dations. The recommendation models of Biased MF andKBTRS have higher error rate on the RMSE and MAE whichmakes the final recommendation list inconsistent. The recom-mendation approaches of DPSOHiK, PCAHTRS, andHSSRS have produced comparative results to the proposedHLTRS.
The efficient utilization of user profiling, context-awaredata, and swarm intelligence algorithms enhance the ratingprediction model of the proposed HLTRS, and the proposedapproach is capable of producing recommendations to bothcold start, and all user sets. The dataset used for experimenta-tion is large and the time taken to produce the final set of
Fig. 3 Comparison of precision for various recommendation approaches
Fig. 4 Comparison of recall for various recommendation approaches
Mobile Netw Appl
recommendations to an active user is also considered to be acrucial factor. With the complex recommendation generationprocess of HLTRS, the proposed approach hasmade a positiveimpact with reliable recommendations generated in the opti-mal computational cost. After determining the relevant set ofPOIs, to generate the final list of recommendations, the pro-posed HLTRS employs conditional filtering approach byexploiting the user’s current location, contextual data, andtravel distance preference. The conditional filtering approachhelps the HLTRS to produce better POI recommendations tothe active user. Based on the target users’ current location andtravel distance preferences, the POI recommendation list is re-ranked and the final top-n recommendations are generatedthrough the user interface. The sparsity of the ratings datamakes a direct impact on the recommendation generation,and scalability of the recommendation model also consideredas an essential feature. By estimating the leaning time taken bythe proposed approach, we found that leaning time raises with
the increasing size of the dataset used. From the experimentalresults, it has been demonstrated that the recommendationapproaches produce accurate recommendation when the finalnumber of recommendations are less and when the number ofthe final top-n list increases it crunches the accuracy of therecommendation list.
From the experiments conducted and the analyses made onthe obtained results, the following conclusions are derived.
& The utilization of multiple swarm intelligence algorithmsis helpful in improving the accuracy of the recommenda-tions generated.
& The proposed HLTRS with utilization of user profiling,context-aware data and trust information has demonstrat-ed its ability to produce reliable POI recommendations.
& The performance of the HLTRS has been improved withthe co-training based rating enhancement approach.
Fig. 5 Comparison of F-Measure for various recommendationapproaches
Fig. 6 Comparison of RMSE for various recommendation approaches
Fig. 7 Comparison of MAE for various recommendation approaches
Fig. 8 Comparison of coverage for various recommendation approaches
Mobile Netw Appl
& HLTRS outperforms standalone approaches and otherbaseline hybrid approaches with the incorporation ofuser’s contextual information.
& The proposed HLTRS has also been executed on the high-performance computing clusters for the generation of top-n POI list and travel planning.
& HLTRS is evaluated for the real-time recommendationscenarios on the mobile environment, and the suggestedrecommendations are user satisfiable and proficient fortravel requirements.
& Useful and new travel recommender system has been de-veloped and presented for the generation of better recom-mendations to meet the changing travel requirements ofthe active target user.
4.3 Recommendations on mobile and evaluation
In the travel context, smartphones help users in assisting themduring the navigational difficulties faced while searching forinformation on a venue or travel route. The travel recommend-er systems on the mobile platform help users by providinginformation on the relevant POIs and available e-services suchas trip planning and mapping services. The mobile-based trav-el recommender systems are found to cost-effective and usersatisfiable over the conventional web-based solutions. As thetravel planning deals with the more than one POI, modelingthe user profile and generating the personalized trip with thechanging contextual information is a complex task. The per-sonalized tour plan has to be designed based on target user’sconstraints such as preferable distance, travel mode and feasi-bility, energy level, accessible location categories, etc. Thetarget user’s constraints with preferences and features of thePOIs are mapped to solve the travel planning problem.
The proposed HLTRS is evaluated on the real time scenar-ios by exploiting mobile recommendation frameworks such asXplorerVU [58], XplorerVU TwB [10], and CHXplorer [12].The mobile recommendation frameworks of XplorerVU,XplorerVU TwB, and CHXplorer are presented in Fig. 9.The proposed HLTRS on the mobile recommendation frame-work generates personalized recommendations based on tar-get users’ interests and preferences. The users interact with therecommendation engine through mobile application’s user in-terface. The co-trained swarm intelligent algorithms are uti-lized on the mobile recommendation framework to producepersonalized POIs and feasible travel plan. The POIs withhigher predicted ratings are re-ranked on the basis of the user’scontextual constraints, and the final set of top-n point of inter-ests are suggested as recommendations to the target user. Toreduce the sparsity and cold start difficulty of the travel rec-ommender system, HLTRS exploits the location-based socialnetwork data to acquire users’ current location, historicalcheck-ins, preferences, contextual information, interests, and
requirements. Based on the input provided, the recommenda-tion set is further refined and the modified recommendationsare generated and provided to the user through the mobileinterface of recommendation framework.
The mobile user interface presents the generated personal-ized POI list with the interactive map feature to provide en-hanced user personalization. All three mobile recommenda-tion frameworks provide a Bsave the plan^ option. And thesesaved trips and plans are considered as user’s explicit prefer-ences and will be exploited for the future recommendationgeneration. The XplorerVU is designed for individual users,XplorerVU TwB is built for a group of users, and CHXploreris developed for recommending cultural heritage locations.The mobile recommendation frameworks have been alreadyproven to be user-friendly with varying users knowledgelevels from the common man to experts. With the user-friendly travel planning process of the proposed HLTRS andproven mobile recommendation frameworks, the travel rec-ommender system produced single destination recommenda-tion, multiple POI recommendation, and travel trip plans. Theuser interfaces of mobile recommendation frameworks areevaluated on the ISO 9241-11-110 standards. The recommen-dation ability of HLTRS on the mobile user interface is eval-uated on three mobile recommendation frameworks with 86participants.With extensive user study on three provenmobilerecommendation frameworks, HLTRS has demonstrated itsimproved performance by providing travel planning and rec-ommendations by incorporating the implicit and explicit pref-erences of the active target users.
4.4 Location-based recommendations for culturalheritage applications
The rapid development of internet technologies andsmartphone applications have reflected with a massive in-crease in the variety of location-based services provided todigital users. With the better utilization of emerging intelligenttechnologies, recommender systems are employed to supportthe travel users visiting the cultural heritage sites. The culturalheritage recommendations differ from the traditional travellocation recommendation as the features and attributes of cul-tural heritages venues requires specific mapping to the users’interests. With the rich information management system spe-cifically designed and developed for the cultural heritage lo-cations, CHXplorer has demonstrated its ability to be an ef-fective mobile decision support mechanism for the culturalheritage travellers. We have employed our proposed HLTRSon the CHXplorer mobile recommendation framework to gen-erate tailored cultural heritage recommendations. The experi-ments are conducted at the cultural heritage sites of theThanjavur city in India for the prediction of the next culturalheritage location and travel plan generation. The experimentalresults and the user study made reveal the enhanced
Mobile Netw Appl
performance of HLTRS on CHXplorer and positive user feed-back on the generated personalized recommendations. Alongwith the travel recommendations to the cultural heritage loca-tions, the CHXplorer also presents the multimedia content onthe user interface with backend support from the dedicatedinformation management system. The mobile interface ofthe CHXplorer is devised to be interactive and user-friendlyfor all types of users and based on user’s interactions; recom-mendation engine is capable of discovering the user’s implicitpreferences. The archaeological cultural heritage sites in theThanjavur, India such as Brihadeeshwara Temple, SchwartzChurch, and Thanjavur Fort Place were considered for theexperimentation of HLTRS on CHXplorer mobile recommen-dation framework to evaluate the developedmodel in real timescenarios. The obtained results revealed the promising perfor-mance of HLTRS for the focused recommendations for cul-tural heritage travelers with enhanced prediction modelthrough exploiting the preferences of the active target user.
5 Conclusion and future directions
The significant progress in the development of recommendersystems has proved their efficiency in handling informationoverloading problem in various application domains. To solvethe user interest prediction and personalization problems of thetravel recommender system, we have developed a new recom-mendation model as Hybrid Location-based TravelRecommender System (HLTRS) to generate the tailor-madepoint of interest recommendations by considering target user’sdynamic needs and preferences. To generate personalized rec-ommendations, HLTRS constructs an individual user profile forpredicting the interesting locations based on the individual andgroup activity information obtained from the location-basedsocial network. The constructed individual user profile
comprises of the explicit and implicit preferences of the targetuser, and it is used in the recommendation generation process.The developed HLTRS utilizes trust and contextual informationof the target user to filter highly relevant POI from the masteritem set resulting in reduced computational complexity. Theensemble-based co-trained swarm intelligence approach ofHLTFS reflects with the enrichment of the user-POI rating ma-trix. The developed HLTRS is designed with the focus to solvethe sparsity and cold start problems of the conventional recom-mender system. The proposed HLTRS is experimentally eval-uated on real time dataset of TripAdvisor for the production ofpersonalized travel recommendations. We have also conductedextensive user study for estimating the capacity of HLTRS forproducing personalized travel recommendations in real timescenarios. The obtained experimental results demonstrate theimproved performance of the HLTRS over the existing base-lines and user study reveals promising recommendation abilityof HLTRS on various mobile recommendation frameworks. Inthe future, we plan to examine the recommender systems re-search for the cross-domain recommendations, and also weintend to study the consensus decision-making problem in thegroup travel recommendations.
Acknowledgements The authors are grateful to the Science andEngineering Research Board (SERB), Department of Science &Technology, New Delhi, for the financial support (No. YSS/2014/000718/ES). Authors express their gratitude to SASTRA DeemedUniversity, Thanjavur, for providing the infrastructural facilities to carryout this research work.
References
1. Bobadilla J, Ortega F, Hernando A, Gutiérrez A (2013)Recommender systems survey. Knowl-Based Syst 46:109–132
2. Leino J (2014) User factors in recommender systems: Case studiesin e-commerce, news recommending, and e-learning
Fig. 9 Mobile recommendationframeworks a XplorerVU [58] bXplorerVU TwB [10] cCHXplorer [12]
Mobile Netw Appl
3. Subramaniyaswamy V, Vijayakumar V, Logesh R, Indragandhi V(2015) Intelligent travel recommendation system by mining attri-butes from community contributed photos. Procedia ComputerScience 50:447–455
4. Subramaniyaswamy V, Logesh R, Chandrashekhar M, Challa A,Vijayakumar V (2017) A personalised movie recommendation sys-tem based on collaborative filtering. International Journal of HighPerformance Computing and Networking 10(1–2):54–63
5. Subramaniyaswamy V, Manogaran G, Logesh R, Vijayakumar V,Chilamkurti N, Malathi D, Senthilselvan N (2018) An ontology-driven personalized food recommendation in IoT-based healthcaresystem. J Supercomput:1–33. https://doi.org/10.1007/s11227-018-2331-8
6. Subramaniyaswamy V, Logesh R, Indragandhi V (2018) Intelligentsports commentary recommendation system for individual cricketplayers. International Journal of Advanced Intelligence Paradigms10(1–2):103–117
7. Vairavasundaram S, Varadharajan V, Vairavasundaram I, Ravi L(2015) Data mining-based tag recommendation system: an over-view. Wiley Interdisciplinary Reviews: Data Mining andKnowledge Discovery 5(3):87–112
8. Zhang Y, Tu Z, Wang Q (2017) TempoRec: Temporal-topic basedrecommender for social network services. Mobile Networks andApplications 22(6):1182–1191
9. Logesh R, Subramaniyaswamy V, Vijayakumar V (2018) Apersonalised travel recommender system utilising social networkprofile and accurate GPS data. Electronic Government, anInternational Journal 14(1):90–113
10. Logesh R, Subramaniyaswamy V, Vijayakumar V, Li X (2018)Efficient User Profiling Based Intelligent Travel RecommenderSystem for Individual and Group of Users. Mobile Networks andApplications:1–16. https://doi.org/10.1007/s11036-018-1059-2
11. Logesh R, Subramaniyaswamy V (2017) Learning Recency andInferring Associations in Location Based Social Network forEmotion Induced Point-of-Interest Recommendation. J Inf SciEng 33(6)
12. Ravi L, Vairavasundaram S, Palani S, Devarajan M (2019)Location-based personalized recommender system in the internetof cultural things. J Intell Fuzzy Syst. https://doi.org/10.3233/JIFS-169973
13. Logesh R, Subramaniyaswamy V (2017) A Reliable Point ofInterest Recommendation based on Trust Relevancy betweenUsers. Wirel Pers Commun 97(2):2751–2780
14. Chen X, Xu X, Huang JZ, Ye Y (2013) TW-k-means: Automatedtwo-level variable weighting clustering algorithm for multiviewdata. IEEE Trans Knowl Data Eng 25(4):932–944
15. Logesh R, Subramaniyaswamy V, Vijayakumar V, Gao XZ, WangGG (2019) Hybrid bio-inspired user clustering for the generation ofdiversified recommendations. Neural Comput &Applic. https://doi.org/10.1007/s00521-019-04128-6
16. Wang GG, Gandomi AH, Alavi AH, Deb S (2016) A hybrid meth-od based on krill herd and quantum-behaved particle swarm opti-mization. Neural Comput & Applic 27(4):989–1006
17. Balusamy B, Karthikeyan K, Sangaiah AK (2017) Ant colony-based load balancing and fault recovery for cloud computing envi-ronment. International Journal of Advanced Intelligence Paradigms9(2–3):204–219
18. Wang GG (2018) Moth search algorithm: a bio-inspiredmetaheuristic algorithm for global optimization problems.Memetic Computing 10:151–164. https://doi.org/10.1007/s12293-016-0212-3
19. Devarajan M, Fatima NS, Vairavasundaram S, Ravi L (2019)Swarm intelligence clustering ensemble based point of interest rec-ommendation for social cyber-physical systems. J Intell Fuzzy Syst.https://doi.org/10.3233/JIFS-169991
20. Guo G (2013) Improving the performance of recommender systemsby alleviating the data sparsity and cold start problems. In Twenty-Third International Joint Conference on Artificial Intelligence
21. Da Costa AF, Manzato MG, Campello RJ (2018) CoRec: a co-training approach for recommender systems. In: Proceedings ofthe 33rd Annual ACM Symposium on Applied Computing (pp.696–703). ACM
22. Blum A, Mitchell T (1998) Combining labeled and unlabeled datawith co-training. In: Proceedings of the eleventh annual conferenceon Computational learning theory (pp. 92–100). ACM
23. da Costa AF, Manzato MG, Campello RJ (2019) Boosting collab-orative filtering with an ensemble of co-trained recommenders.Expert Syst Appl 115:427–441
24. Ravi L, Vairavasundaram S (2016) A collaborative location basedtravel recommendation system through enhanced rating predictionfor the group of users. Computational Intelligence andNeuroscience 2016:7. https://doi.org/10.1155/2016/1291358
25. Goldberg D, Nichols D, Oki BM, Terry D (1992) Using collabora-tive filtering to weave an information tapestry. Commun ACM35(12):61–71
26. Konstan JA,Miller BN,Maltz D, Herlocker JL, Gordon LR, Riedl J(1997) GroupLens: applying collaborative filtering to Usenet news.Commun ACM 40(3):77–88
27. Herlocker JL, Konstan JA, Terveen LG, Riedl JT (2004) Evaluatingcollaborative filtering recommender systems. ACM Transactionson Information Systems (TOIS) 22(1):5–53
28. Sankar H, Subramaniyaswamy V, Vijayakumar V, Kumar SA,Logesh R, Umamakeswari A (2019) Intelligent sentiment analysisapproach using edge computing-based deep learning technique.Software: Practice and Experience. https://doi.org/10.1002/spe.2687
29. Subramaniyaswamy V, Logesh R, Abejith M, Umasankar S,Umamakeswari A (2017) Sentiment Analysis of Tweets forEstimating Criticality and Security of Events. Journal ofOrganizational and End User Computing (JOEUC) 29(4):51–71
30. Malathi D, Logesh R, Subramaniyaswamy V, Vijayakumar V,Sangaiah AK (2019) Hybrid Reasoning-based Privacy-AwareDisease Prediction Support System. Comput Electr Eng 73:114–127
31. Lathia N, Hailes S, Capra L (2008) Trust-based collaborative filter-ing. In: IFIP international conference on trust management (pp.119–134). Springer, Boston
32. Massa P, Avesani P (2009) Trust metrics in recommender systems.In: Computing with social trust (pp. 259–285). Springer, London
33. Kant V, Bharadwaj KK (2013) Fuzzy computational models of trustand distrust for enhanced recommendations. Int J Intell Syst 28(4):332–365
34. Guo G, Zhang J, Thalmann D, Basu A, Yorke-Smith N (2014)From ratings to trust: an empirical study of implicit trust in recom-mender systems. In: Proceedings of the 29th Annual ACMSymposium on Applied Computing (pp. 248–253). ACM
35. Gupta S, Nagpal S (2015) Trust aware recommender systems: asurvey on implicit trust generation techniques. InternationalJournal of Computer Science and Information Technologies 6(4):3594–3599
36. Ujjin S, Bentley PJ (2003) Particle swarm optimization recom-mender system. In: Proceedings of the 2003 IEEE SwarmIntelligence Symposium. SIS'03 (Cat. No. 03EX706) (pp. 124–131). IEEE
37. Katarya R, Verma OP (2016) A collaborative recommender systemenhanced with particle swarm optimization technique. MultimedTools Appl 75(15):9225–9239
38. Choudhary V, Mullick D, Nagpal S (2017) Gravitational searchalgorithm in recommendation systems. In InternationalConference on Swarm Intelligence (pp. 597–607). Springer, Cham
Mobile Netw Appl
39. Bedi P, Sharma R (2012) Trust based recommender system usingant colony for trust computation. Expert Syst Appl 39(1):1183–1190
40. An J, Kang Q, Wang L, Wu Q (2013) Mussels wandering optimi-zation: an ecologically inspired algorithm for global optimization.Cogn Comput 5(2):188–199
41. Bouyer A, Hatamlou A (2018) An efficient hybrid clustering meth-od based on improved cuckoo optimization and modified particleswarm optimization algorithms. Appl Soft Comput 67:172–182
42. Devarajan M, Ravi L (2018) Intelligent cyber-physical system foran efficient detection of Parkinson disease using fog computing.Multimed Tools Appl. https://doi.org/10.1007/s11042-018-6898-0
43. Vijayakumar V, Malathi D, Subramaniyaswamy V, Saravanan P,Logesh R (2018) Fog computing-based intelligent healthcare sys-tem for the detection and prevention of mosquito-borne diseases.Comput Hum Behav. https://doi.org/10.1016/j.chb.2018.12.009
44. Gavalas D, Kasapakis V, Konstantopoulos C, Pantziou G, Vathis N,Zaroliagis C (2015) The eCOMPASS multimodal tourist tour plan-ner. Expert Syst Appl 42(21):7303–7316
45. Cenamor I, de la Rosa T, Núñez S, Borrajo D (2017) Planning fortourism routes using social networks. Expert Syst Appl 69:1–9
46. De Pessemier T, Dhondt J, Martens L (2017) Hybrid group recom-mendations for a travel service. Multimed Tools Appl 76(2):2787–2811
47. Logesh R, Subramaniyaswamy V, Malathi D, Senthilselvan N,Sasikumar A, Saravanan P (2017) Dynamic particle swarm optimi-zation for personalized recommender system based on electroen-cephalography feedback. Biomed Res 28(13):5646–5650
48. Brilhante IR, Macedo JA, Nardini FM, Perego R, Renso C (2015)On planning sightseeing tours with TripBuilder. Inf Process Manag51(2):1–15
49. Kurata Y, Hara T (2013) CT-planner4: Toward a more user-friendlyinteractive day-tour planner. In: Information and communicationtechnologies in tourism 2014 (pp. 73–86). Springer, Cham
50. Logesh R, Subramaniyaswamy V (2019) Exploring HybridRecommender Systems for Personalized Travel Applications. In:Cognitive Informatics and Soft Computing (pp. 535–544).Springer, Singapore
51. Subramaniyaswamy V, Logesh R (2017) Adaptive KNN based rec-ommender system through Mining of User Preferences. Wirel PersCommun 97(2):2229–2247
52. Zhang M, Tang J, Zhang X, Xue X (2014) Addressing cold start inrecommender systems: A semi-supervised co-training algorithm.In: Proceedings of the 37th international ACM SIGIR conferenceon Research & development in information retrieval (pp. 73–82).ACM
53. Zhang Q, Wang H (2015) Collaborative Multi-view Learning withActive Discriminative Prior for Recommendation. In: Pacific-AsiaConference on Knowledge Discovery and Data Mining (pp. 355–368). Springer, Cham
54. Quang TN, Phuong ND (2015) Collaborative filtering by co-training method. In: Knowledge and Systems Engineering (pp.273–285). Springer, Cham
55. Matuszyk P, Spiliopoulou M (2017) Stream-based semi-supervisedlearning for recommender systems. Mach Learn 106(6):771–798
56. Logesh R, Subramaniyaswamy V, Malathi D, Sivaramakrishnan N,Vijayakumar V (2019) Enhancing Recommendation Stability ofCollaborative Filtering Recommender System through Bio-inspired Clustering Ensemble method. Neural Comput & Applic.https://doi.org/10.1007/s00521-018-3891-5
57. Younus A, O’Riordan C, Pasi G (2014) A language modeling ap-proach to personalized search based on users’ microblog behavior.In: European Conference on Information Retrieval (pp. 727–732).Springer, Cham
58. Logesh R, Subramaniyaswamy V, Vijayakumar V, Gao XZ,Indragandhi V (2018) A hybrid quantum-induced swarm intelli-gence clustering for the urban trip recommendation in smart city.Futur Gener Comput Syst 83:653–673
59. Dorigo M, Birattari M, Stützle T (2006) Ant Colony Optimization-Artificial Ants as a Computational Intelligence Technique. IEEEComputational Intelligence Magazine
60. Yang XS, Deb S (2009) Cuckoo search via Lévy flights. Nature &Biologically Inspired Computing. In 2009 NaBIC 2009 WorldCongress on: IEEE (pp. 210–214)
61. Adomavicius, G., & Tuzhilin, A. (2011). Context-aware recom-mender systems. In: Recommender systems handbook (pp. 217–253). Springer, Boston
62. Deshpande M, Karypis G (2004) Item-based top-n recommenda-tion algorithms. ACMTransactions on Information Systems (TOIS)22(1):143–177
63. Koren Y, Bell R, Volinsky C (2009)Matrix factorization techniquesfor recommender systems. Computer (8):30–37
64. Vijayakumar V, Vairavasundaram S, Logesh R, Sivapathi A (2019)Effective Knowledge Based Recommender System for TailoredMultiple Point of Interest Recommendation. International Journalof Web Portals (IJWP) 11(1):1–18
Publisher’s Note Springer Nature remains neutral with regard to jurisdic-tional claims in published maps and institutional affiliations.
Mobile Netw Appl